🔍 Read the full analysis: 722 Proofs, One Question: Can OpenAI’s AI Mathematics Lead Anywhere? on ThorstenMeyerAI.com
Get movie-night favorites delivered free — and shop member deals
- Fast, free delivery on millions of items
- Access to Prime Big Deal Days deals on October 6–7
- Prime Video, Amazon Music and more included
TL;DR
OpenAI says an unnamed, unreleased model produced 722 mathematical manuscripts across 372 families of results, selected from about 4,000 problems. The claims include work on major open problems, but OpenAI says outside mathematicians have not confirmed them; the collection’s value will depend on verification and whether researchers can use its methods.
OpenAI published 722 mathematical manuscripts on Monday, presenting results produced by an unnamed, unreleased model and selected from roughly 4,000 problems. The collection includes claims involving major open questions, but the company and its repository caution that the work has not been confirmed by outside mathematicians.
The manuscripts are arranged into 372 families of related results and cover areas including number theory, geometry, operator algebras, topology, theoretical computer science and mathematical physics. OpenAI says an average result took about three hours of ChatGPT Pro thinking compute. The collection is available under an Apache-2.0 license.
Among the claims are a proof of the Unique Games Conjecture, a resolution of Hilbert’s tenth problem over the rationals, and results concerning nonabelian free group factors, the Riemann zeta function, the Hodge conjecture for CM abelian varieties and Mahler’s conjectures. These are claims in the manuscripts, not independently established solutions. OpenAI’s repository says some results are not formally verified and warns that “some of the unformalized results could have issues.”
OpenAI says it selected the published work from around 4,000 problems based on what it considered an appropriate level of significance. The company supplied ten abridged reasoning summaries for the 372 families. Many results have Lean formalizations, but not all. The Riemann zero-free-region manuscript was edited by people for readability, and OpenAI describes that result and the Hodge result as exceptions to its standard process.
722 proofs, one question: will any of OpenAI’s AI mathematics actually lead anywhere?
An unreleased, unnamed model produced claimed proofs of results that would each define a career. Sam Altman calls them “claims not yet confirmed by outside mathematicians.” The real question isn’t whether it’s impressive. It’s whether answers nobody understands become discoveries anyone can build on.
Same day: Alon, Bloom, Gowers, Litt, Sawin post a digested, human-verified version. The model for success.
Connes rigidity counterexample challenged within a day — constructed groups fail the required condition. Three rival machine “counterexamples” from different labs now circulate.
~10,000 agents, 88 hours, est. ~$22M at retail. Priority dispute; 25 Fields Medalists sign “A Severe Misalignment” — not saying it’s wrong, saying it’s not understood.
Altman now hedges at announcement — a shift from September. Verification has barely started.
Humans extract the technique, write it up, build on it. This is where downstream discovery comes from.
The question is answered; nobody learns anything reusable. Closes a door without opening a field.
The proof breaks, or proves a statement that doesn’t match the conjecture as mathematicians mean it.
The Unique Games Conjecture is the clearest case. Results like the optimality of Goemans–Williamson for Max-Cut are proved assuming UGC. A correct proof converts them all — no understanding required. A zero-free strip for zeta works the same way for prime-distribution results. Free group factors, Kadison, Mahler would redirect whole programmes — but how depends on the method, which means digestion.
Technology. A Navier–Stokes blow-up proof doesn’t change how anyone designs aircraft; engineering turbulence models never depended on the answer. Near-term consequences are mathematical, not industrial. “AI will cure cancer next” skips several steps.
“Verification abundance, adjudication scarcity” — making proof-checking cheap doesn’t reduce the burden of deciding what’s true and what matters. 722 manuscripts land on a review system built for a trickle, filtered by a selection nobody outside OpenAI made.
Humans re-deriving results, like Alon–Gowers et al. in May
Other people’s work building on these manuscripts
How many unformalized results survive expert checking
Do the Lean statements match the real conjectures?
Do any survive peer review?
Some of it, yes — where a literature is waiting (UGC), a correct proof pays off immediately; where a proof carries a new technique humans digest, it can open a field. Most of it, probably not on its own: at 722 manuscripts with 10 reasoning summaries, the Four Colour pattern is the likely default unless mathematicians are funded and given time. And some will be wrong — OpenAI says so itself. It’s an industry pattern, not one company’s: the forced-Euler result came from an Anthropic researcher, and rival machine-generated Connes “counterexamples” circulate from different labs. The proofs arrived this week. The discoveries, if they come, will arrive at the speed of human understanding.
Verification Will Shape the Payoff
The release matters less as a tally of claimed solutions than as a test of whether AI-generated mathematics can produce work other researchers can verify and extend. A correct proof may settle a question; a useful mathematical discovery can also provide methods, concepts or techniques that open new lines of research.
The Unique Games claim illustrates the possible stakes. The conjecture underpins a substantial body of theoretical computer science about the limits of approximation algorithms. If a proof withstands scrutiny, researchers could revisit results that depend on it. But until independent mathematicians check the argument and determine what it establishes, those consequences remain conditional, not confirmed.
There is also a practical challenge: hundreds of manuscripts can take substantial expert time to assess, particularly when proofs are long or hard to interpret. The central measure of impact will be whether mathematicians can extract and reuse the underlying ideas, rather than simply whether a model can generate answers to difficult questions.
As an affiliate, we earn on qualifying purchases.
A Mixed Record This Year
This is OpenAI’s fourth major mathematics release this year, according to the source account. In May, a model produced a counterexample to the Erdős unit-distance conjecture. Five mathematicians published what they called a digested, human-verified version the same day, illustrating a process in which people turn machine output into an argument that the field can evaluate.
An August release called “Ten Advances” had mixed results: a claimed counterexample to Connes’s rigidity conjecture was challenged within a day. Critics said the constructed groups did not meet a condition required by the conjecture. In September, OpenAI announced a Lean-formalized Navier–Stokes result produced using about 10,000 concurrent agents over 88 hours. That announcement prompted a dispute over research priority and criticism from 25 Fields Medalists, who argued that using famous problems as benchmarks without human understanding can conflict with mathematics’ purpose.
Those episodes do not establish whether the new manuscripts are correct. They do show why verification and interpretation are separate tasks: formal checking can help establish that a proof follows specified rules, while experts still need to judge what the result means and whether it advances the field.
formal verification tools for mathematicians
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Independent Checks Remain Pending
The source material does not report independent confirmation of the new collection’s major claims. It is not yet clear which manuscripts will survive expert review, whether their formalizations cover the central arguments, or how much of the work can be checked with the available summaries. OpenAI’s selection of problems and results also means the published set is not an independent survey of everything the model attempted.
Even if individual proofs prove correct, their broader value is unsettled. Researchers will need to determine whether the arguments contain reusable ideas, merely settle particular statements, or rely on errors or mismatches between a claim and the problem mathematicians intended to solve.
mathematical proof assistant software
As an affiliate, we earn on qualifying purchases.
As an affiliate, we earn on qualifying purchases.
Mathematicians Must Test the Claims
The next step is independent mathematical review of the manuscripts, including checking the statements, proof details and any Lean formalizations. Researchers will also need to compare the claims with the precise versions of the problems they address and document any corrections or successful verifications.
No timetable for that review is given in the source material. The clearest measure of what follows will be whether mathematicians produce readable, verified accounts and then use the methods in further work. Until that happens, the collection is a substantial set of AI-generated claims under examination, not a confirmed catalogue of new mathematical discoveries.
As an affiliate, we earn on qualifying purchases.
Key Questions
What did OpenAI publish?
OpenAI published 722 mathematical manuscripts, grouped into 372 families, and attributed them to an unnamed, unreleased model. The work was selected from roughly 4,000 problems.
Have mathematicians verified the claimed proofs?
The source material says the claims have not yet been confirmed by outside mathematicians. OpenAI’s repository also warns that some unformalized results could have issues.
What are some of the biggest claims?
The collection includes claimed results on the Unique Games Conjecture, Hilbert’s tenth problem over the rationals, the Riemann zeta function and the Hodge conjecture for CM abelian varieties. Their appearance in the collection does not mean they have been accepted as correct.
Why does a correct proof not automatically mean a major advance?
A proof can settle a question without giving researchers a method they can reuse. The wider impact will depend on whether mathematicians can understand, verify and build on the arguments.
Source: ThorstenMeyerAI.com
Fall Picks
fall essentials
As an affiliate, we earn on qualifying purchases.
